[time-nuts] GPSDO uncertainties

Bob kb8tq kb8tq at n1k.org
Mon Oct 8 11:11:00 EDT 2018


The question becomes very much, what confidence level are you after? Is 99%
good enough? Are you after 99.9999%? Obviously the stability gets worse as 
you crank on more nines.

Next layer to the onion is - how long have things been running and under what 
conditions are they running? A few days is probably the minimum for any sort of 
reasonable performance. A month or more will indeed help a bit. A stable temperature
environment is also going to have a direct impact on performance. 

Things like survey position and antenna location ( number of sats viewed at all times)
will have a more subtle impact on frequency. Solar storms and ionospheric stability (sun
spot cycle) eventually get into the picture if you put in enough nines. 

So simple answer: better than 2x10^-9 with any practical number of nines involved. Better
than 2x10^-10 with rational numbers (99.9 or 99.99). Maybe 2x10^-11 at the 99% level, but
probably not.  

Since this is a very non-standard measure your numbers may not match my numbers. ADEV 
was invented primarily because this sort of measure has so many issues. One of the issues 
is that you need a *lot* of data to even begin to estimate the confidence levels and they are
very dependent on data set size. A typical spec tends to have weird stuff like “99.9% 
based on 4 out of 5 runs of 100 hours”.  Statements like that tend to make a statistics guy
groan ….(or throw something at you).


> On Oct 7, 2018, at 9:59 PM, donald collie <donaldbcollie at gmail.com> wrote:
> I have a GPSDO which has a Trimble receiver, and Trimble OCXO inside.[I
> think it runs Lady Heather] What I would like to know is what uncertainty
> the 10MHz output will be with a 10 second count.This presumes that the
> GPSDO is receiving satellite signals, and the OCXO is being suitably
> diciplined. Should I expect 1 part in 10E12?, 1 part in 10E10?. This device
> gives superior performance to an OCXO barefoot - but how much better? Would
> somebody in the group who knows the answer, please wax elequently.
> Thankyou,.................................................................................................Donald
> _______________________________________________
> time-nuts mailing list -- time-nuts at lists.febo.com
> To unsubscribe, go to http://lists.febo.com/mailman/listinfo/time-nuts_lists.febo.com
> and follow the instructions there.

More information about the time-nuts mailing list